MP3 player for MSX

Is the size of the data buffer configurable or is it always 3.5kB? @128kbps, the buffer is around 200ms if I calculated correct. This is great in an mp3 player but maybe a bit too long if you want do do sound effects that are synchronized with video or user input.
On the other hand I'm not sure if mp3 is a good format for sound effects. In a game its probably better to have high quality background music in mp3 and do sound effects with another sound chip and then the delay doesn't matter.

Having a configurable 'buffer near empty' flag is not needed I think but could be good. But I guess it need to be set earlier than 512 bytes before buffer empty, so it can be used in bit rates > 192kbps. If its fixed I think 1024 bytes is a good number.

Is this VS1011 doable as not-so-heavy addition for the OCM?
You mean integrated in the FPGA, right?I think it may be a bit heavy. Not sure how many gates it would need but it sounds like there aren't too many free ones in the ocm.

With mp3 serving for music, you can drop FM-Pac, MSX-Audio, SCC and PSG at once (well, maybe a small simple soundgenerator for some sfx)

You don't need internal memory, if you "stream" it from hard disc. I don't like the idea of having memory on the MP3 card too much.
- big delays before playing an MP3
- additional costs
- nothing else than an "iPod" controlled by the MSX
If my calculations are not totally wrong, you will still have 75% CPU time and nearly the complete MSX memory available while playing 128kbps MP3s from HD/CF card.

Is the size of the data buffer configurable or is it always 3.5kB?

That doesn't matter. It's your choice how much bytes you load into the buffer after receiving the "nearly empty" signal. Regarding the reaction time and "realtime" sound-effects: The question is, if the MP3 card does nothing at the moment, and I start to fill the buffer: After what delay will the MP3 decoder throw out the decoded audio? I think that has nothing to do with the size of the buffer, has it?