Hallo!
I just tested it. You wrote "this is very experimental and may crash your patch". ... and yes, it did!
;)
Unfortunately when it doesn't crash the patch it also interupts audio. BTW you should add the option of using the "-resize" in the load message so you can use the same commands as for soundfiler without having to change all the load messages when replacing "soundfiler" with "sndfiler". I guess "sndfiler" should simply ignore "-resize".
The version in cvs (or now svn) supports that - I guess you got the one from my homepage ? The sndfiler should not crash pd - then something is definitely wrong. Which OS are you on ? Maybe there is somewhere a library problem and I suggest you to compile it yourself, then there shouldn't be a crash.
Also if it works as it should there could be still some audio dropouts, because the dsp graph has to be recalculated in pd (this was discussed quite some times on this list) and there is no (easy) way to solve this problem. However, I used sndfiler in some performance patches without dropouts.
I need something extremely stable as I am developing a hardware instrument (sample player) that runs with pd. Anything unstable is absolutely out of the question. I guess I'll have to keep loading the patches as the get recalled instead of loading in the background after the instrument stared playing already.
If it helps you can also consider the iem16 library, then you can load more samples into RAM ... ;)
LG Georg