Ah, it looks like this is related to high speed modem port modifications? I haven't looked in detail at how those work. I was under the impression that some of those modifications use a spare output line (ACIA RTS or YM chip bit 7) to enable the higher speed clock. Are some people just putting a PLL in between timer D and TX/RX clock?
I'm not really sure why Hatari has a weird clock, but I think the modem port implementation in Hatari is incomplete.
It seems like the best way to get a higher speed modem port is to use an MCU to do clock extraction on the incoming data. That would allow you to go up to about 1 megabaud without overclocking the MFP. I'm not sure if anyone has done such a modification?
There's another possible modification which involves running the 500kHz ACIA clock to the MFP, allowing MIDI input and output via the modem port. I suppose MINT would also detect this as a weird clock.
I did some tests to measure interrupt latency on MINT, and to my surprise the figures were barely any worse than standard TOS. The ACIA interrupt handler is still as slow as TOS, so a MIDI sequencer would still need to trap that routine and replace it. My test program used timer A, seemingly without upsetting MINT. I'm not sure if timer B is available for use.