Unfortunately headphones cannot be designed to have opposite phase, combfilter or FR compensation to compensate for the many different types and possible settings of various X-feeds.

X-feed may give some or a slight improvement with some recordings in the spatial aspect though also depending on the individual.
The holy headphone audio grail it will never be.

With some headphones that are somewhat tonally skewed by themselves in the opposite direction as what the X-feed does it may also bring some relief.
In that sense I can go along with the FR needing to be different from the ones that are perceived as accurate under normal use.

Headphones are headphones with their upsides and downsides, speakers are speakers with their upsides and downsides.
One has to learn to live with them, they can never substitute each other but may complement each other depending on the purpose they are used for.

I agree for the most part, but I still feel as if headphone crossfeed is a worthy endeavor. On certain recordings with too much stereo separation, listening on headphones is undoable. As in it's literally painful to hear a drum solo rattle off in one of my ears. Even a basic crossfeed circuit gets rid of that obviously, but introduces too many other problems for me to consider it high fidelity.

At this point, we're (not me) just making incremental improvements to getting that last bit of transparency. The better fix, of course, would be to buy into some nice loudspeakers   

VNoPhones is a standard crossfeed, like Xnor's, but it doesn't have any highs/lows strength processing. Combs mono signal.
Mine should be audibly superior, as it doesn't feed center signals.

About 283us setting, that's almost the default in mine now (or very close) - it is 60 degrees for about 15 cm ear spread according to the spherical model. I've set it to 14 cm, which fits my head better.
25% is -12 dB at bass, which is slightly less crossfeed than I set mine to by default. (-9 dB, would be ~33%)

The Advanced Crossfeed also simulates center summing of speakers, which other crossfeeds don't do.

--
The "real fix" would be buying or building loudspeakers, acoustically treating the room, getting rid of the neighbors.

Full size: Hifiman HE-500 w/ jergpad clones, w/o grille and grille cloth on both sides; equalized. Amp: neutrally-configured Eddie Current Super 7. Source: Presonus FP10. (I might one day do this with equalized Paradox too.)
IEMs: VSonic GR07, modded equalized Sony MH1. (Soon, Brainwavz B2 and Hifiman RE-400 again). Amp: Leckerton UHA-6s mkII w/ ADA4898-1. Source: Presonus FP10.
Speaker comparison: Tonsil Space '86 in near field, Amp: some Tonsil, Source: Leckerton UHA-6s mkII w/ ADA4898-1
(Soon, Lynx HiLo will become the source for everything except the speakers, which aren't at my place.)

Now it doesn't try to simulate speakers "too much" anymore. I bet you've tried that *bad* nonlinear v3 I've released earlier - it was way broken.
Should be much cleaner than VNoPhones too, the defaults are close. If you set the HF feed and LF feed to -12 dB and head size to 15 cm, center boost to 3 dB, they should be nearly identical in terms of soundstaging.
What might be happening is that the combing you get from the VNoPhones actually equalizes your headphones.