It looks like an old bug has resurfaced - indeed pvoc should resynth to the original sample size/type. I will make sure this is included in the next release.
Regarding the mono question - this is not so easy to change! For long historical reasons the current .ana file format is actually based on WAVE, and the channel field is currently used to indicate the size of the analysis frame (i.e. channels= FFt bins), so there is no means of defining a stereo or N-channel analysis file.
My solution-in-progress has been to define a new PVOCEX analysis file format (extension .pvx), which supports N-channel data and is now the standard format in Csound. Integrating this into CDP is unfortunately a "non-trivial" task; not least as the question also arises in regard to all the other frequency domain CDP file types- transposition, etc. Really. new file formats are needed for these too. And then the code for all the spectral processes would need to be updated (where appropriate) to handle multi-channel data. In the meantime, the current version of "pvplay" will play back a stereo .pvx file.
Current recommended practice (in some cases managed behind the scenes in e.g. SoundShaper) is to split multi-channel files into mono files (use channelx from the Toolkit or "housekeep chans"), process each using pvoc as desired, and then re-interleave the outputs using "submix interleave" or "interlx". The same applies to (for example) the distortion programs, as zero crossings are unlikely to coincide across channels.
That said, multi-channel pvoc processing presents in any case certain technical problems in that as soon as any frequency transformations are applied (including pitch-=shift or time-scaling), the original stereo image is usually destroyed as the subtle inter-channel phase relationships get knocked around. Thus in the majority of cases, pvoc is idiomatically a "mono" sound-design process, with successful (or useable) multi-channel results very much dependent both on the nature of the input and on the trasnformation used.