As you should be aware, there is already a time stretcher available in the sample editor which essentially does what you want, but not in realtime. However,
1) We are not in control over the algorithm used by the time stretcher, as this is an external library. Any future update of the library might change the output of the time stretcher, meaning that if it could be used in a real-time scenario, modules would no longer sound the same.
2) the quality of un-patented time stretchers that are also compatible with OpenMPT's license (this excludes Rubberband) is rather low and not something I would actually consider to be worth using in a realtime effect.

Yes, I'm having to use the TimeStretcher a lot, but when I need the same sample at different tempos, using multiple copies of basically the same sample (some longer than others) increases the file size, sometimes by a lot.

I understood the quality would be questionable, but if real-time time-stretching can be done at once (apart from the player), I've asked myself how difficult it would be to apply on demand. The player i think would have to store a temporary copy of the sample (the part after the command call), reconfigure it, then apply it at the right time.
The FR is only recorded for future devs who might want to tackle this challenge, when the technology and perhaps MPT's code is more amenable to such a function.

The thing is that a perfect time stretcher is never going to exist - laws of physics prohibit that. There can only be really good approximations (Rubberband being one of them, as far as I'm aware), but the only serious, license-compatible library I am aware of that does time stretching is the one we use, and it doesn't sound particularly great to my ears.