I remember reading some notes about using crossover components as close as possible to drivers.

2354 is vented ,I guess you cannot use it in Silicone.
For mid range you can use 1700 headspace is good.

You will have to use dampers to smooth the peaks.

Here's one VSonic experiment. Nicely linearizes them by shifting resonances down a lot. Plus it makes them fit very well and look ridiculous in ears, like Frankenstein monster's bolts.
The shifted resonances nicely linearize 1k - 3k region with slight bump at 4k. (And they're not at the annoying 6.5k ear canal resonance.) About 3 dB differences.
Bass response is also shifted down, while extreme highs extension is a bit improved.
Sensitivity is reduced 12 dB or so. (toggling 4x gain on UHA-6 gives same loudness)

The tips feel better in the ear for some reason, perhaps because now I don't get to jam them super deep and they fit precisely on the first bend. Not like a silicone CIEM, but still very good.

How to make these hacked tips? Jam a 2 cm long silicone 4mm tube down Hifiman long tips using tweezers, making it into a long tip with 1 cm stem and internal diameter 1mm smaller. (It's 4mm, GR07 is made for 3mm tips but can hold some 4mm too.)

This makes me yet more hopeful for the CIEM experiment and custom tips with these.

--

I will be sculpting and building an ear-and-cheek attachment for the measurement microphone (with some sculpted ear canal) and trying out a new excellent silicone foam process. That foam has interesting properties:
- hard until the last step, making for machining as in pure silicone
- flows exactly like pure silicone
- open cell, but not a huge lot
- soft with some memory but mechanically resistant - softer than Shure olives, firmer than Comply (will check with various source silicone; this is with one generic junk)
- smooth surface
- shock resistance and low resonance, usable for a damping material
- no need for annoying gas tanks or toxic, smelly solvents

The main drawback is that the process requires a full crosslinking step. (When it's machine-able.) Me and a friend will be trying to optimize it to not have to wait 24 h and dry it some more afterwards.
This will be an interesting material to make a CIEM and tips in.

Another fine day of hacking.

Highlights: learned to use the mechanical impression gun, figured out how to make a negative properly, figured out tube routing (but not tried yet).
Another round of silicone foam experiments - previous were a failure probably due to gloves containing sulphur.
Lots of practice on placing foam stops - now no leaks.

Thoughts and hints:

• Impression material makes a huge difference. Warner Tech-care Quick-Pro ("Clear" color) is the worst. Out of 4 impressions, all had major flaws. I think it is too viscuous and rips while applying due to sticking to the gun. Siemens Silhouette singlets are tricky to use - have to mix it very fast and apply, tends to harden too quickly in the syringe. Usually leaves tiny "fold" artifacts as it is slightly too viscuous when applying. (Blue color) Westone SiliClone is the best. Both impressions made with it were perfect and it is very efficient to boot. I think like Silhouettes it also expands, but is less viscuous. (Pink Color)
• You have to lacquer or apply release compound for silicone-silicone mold - that wasn't indicated in Smooth-On Mold Star 16 Fast instructions. Wasted one pair of impressions on this.
• The foam stop has to be placed uncomfortably deep and getting the end of the syringe there is notably painful and uncomfortable.
• The impression gun needs an improvement to be able to use a whole DS-50 cartridge - it leaves about 1.5 cm in it.
• Applying the foam stop yourself is quite easy. I think it is much harder to put one to somebody else's ear canal, even with an otoscope and/or correct endoscope.
• Lacquer coat is best done by dipping a whole impression in it. Have to really attach it to something though so that it is easy to get out.

I've made some new pictures and also a movie, but it is missing the foam stop application step. I'll make a better one later, anyway there are plenty of those on the net, it's not anything special.

Grrr, I had to post this multiple times, got a white screen. The remaining pics are available here on Dropbox. Unfortunately I can't link to them directly, or they don't get the nice thumbnails and take the whole page. 
Also a friend's "gypsum bust" project: pics on Dropbox.

Doing the impressions yourself is part easier, part harder. The easier part (if you have a steady hand) is placing the foam stop correctly - you will feel it expand and be very uncomfortable at the right depth.
The hard part is actually applying the impression material - the position, especially with the gun, is highly unnatural, as if you were touching a shotgun to your forehead from the side - and you also need to exert non-neglible amount of pressure on the release.

About the bust thingy - what the friend was doing here is the following:

• Take a Guy Fawkes' mask
• Build a polyclay (non-sulphur plasticine maybe?) sculpture of the same shape below it
• Put the mask on it and fill any missing parts, cut any excess as necessary
• Cool it down so that the clay hardens
• Apply silicone with thixotropic additive and some light petrol solvent - the reason for the solvent is that the silicone was very old. Acetone didn't work.

Later plan likely involves filling the resulting silicone mold with gypsum, we'll see.
I'm not entirely sure why he used silicone there for any reason, I'll ask again.