It'd be nice if the MGET response could be multiplexed somehow - I don't
know how much overhead that would add, but it would have the benefits of
1) (as Steinar wants and the prime benefit of Mcom's parallel download)
page layout can happen much much sooner.
2) smaller images complete downloading sooner, so if the download were to
hang halfway (as connections oftentimes do) the chances are greater that
more objects would have been completely obtained (the flip side is that
it also increases the chances that you have half-objects)
There are enough benefits from the current MGET proposal to not make this
necessary in my mind, but it'd be nice if the possibility for this was
always left open for the next HTTP rev.