There are two types of parallel observations: Coordinated Parallel and Pure Parallel.
Coordinated Parallel means part of the same proposal, so it has to have the same scientific justification: you have to come up with a good case for why the different observations would be useful for the same project. (This is case with CANDELS.)
Pure Parallel means something independent of the main observation, which I think is more what you're asking about. The proposal guidelines currently note that these are allowed only when the primary observation is spectroscopy (with the COS or STIS instruments).
One factor is that you need some specification for how you use a given instrument: what filter, what exposure time, what spectroscopic grating, etc. This requires that someone consider a plausible science case for what are effectively random pointings. (And you need to make sure that you're not accidentally going to try taking a long exposure of Vega, which might actually damage an instrument.)
And, yes, I gather there are data-storage and overhead limits, as well as the idea that running the data transmitters at maximum may shorten their lives. From the WFC3 Instrument Handbook (e.g., here):
The primary restriction on parallel observations, both coordinated and
pure, is that they must not interfere with the primary observations:
they may not cause the primary observations to be shortened; and they
must not cause the stored-command capacity and data-volume limits to
be exceeded.
and
In order to prolong the life of the HST transmitters, the number of
parallels acquired during each proposal cycle is limited. Proposers
must provide clear and strong justification in order to be granted
parallel observing time.
(I think the practical answer to your "how many instruments" is two, though I haven't seen an explicit statement to that effect. This page includes specific limitations for individual instruments -- e.g., which ones can be used in Pure Parallel mode and with what other instruments.)
