I'm pretty sure that config delays the whole BOINC startup, it doesn't stagger start. So while it will be slightly helpful so the OS/other programs can get their initializing done, it's not a silver bullet to the WU startup problem.

I can provide some real world SSD usage input. In early 2013 I switched to an Intel 64 Gb X25-E SSD for the Boinc data directory. I have been running 16 concurrent CEP work-units since this time, and the MWI is still at 96%. I over provisioned the drive by 50%, and it has ~580 Tb of writes. These are solid SLC drives and can be found cheap on ebay, etc.

I have also had very good durability with some early Intel SSDs. Given how cheap SSDs are these days, you can probably just purchase a good Crucial or Samsung and it will last a long time. Just set the drive space to what you need and over provision the rest for enhanced durability. It's also much cheaper and reliable to go this route verses a RAM disk.

There's 4 elements when running totally in RAM:

1) The model as being processed in RAM
2) The checkpoint/temp files/slot data
3) The BOINC project storage.
4) The swapfile/VM [if that's relevant]

Oh dear, where does it save if you hibernate such a device ;?

On RAM/VM use, BOINCtasks keeps the peak RAM/VM data in history and I just sorted that history on RAM use. A C22 molecule did peak out at 1.9GB [yes you read that right], and VM size matched that. In November had several that did 1.5GB, C26 and C28. The bulk seems to need 300-400Mb, VM similar, so betting on 170MB RAM and 700MB VM [saw those somewhere up in this thread] is asking for trouble i.e. will not be surprised with allowing a serious overage in the sizing, or results will go south by south west. The bigger the mol, indicated by the Cnn relationship, the greater the needs. Why 2.5GB space is checked for during work requests is a good sign of the possible maxes encountered.

I have to agree with SekeRob* about the memory requirement since after writing the post in which I pointed out the 170+700 MB memory size requirement I've seen some WUs terminated at 4th job (credit still granted) for a lack of memory, they required up to 1,7 GB in my case.
Although when I was gradually increasing from 2 to 8 concurrent WUs reserving more memory to the BOINC folder only when a WU errored out for lack of memory I found out that increasing the swap partition allowed to increase the numer of concurrent WUs without increasing memory reserved.

So no swap isn't a good idea, JTG@[H].

Data persistence in case of power failure, crash, etc. I know you can add protection to a hardware Ram disk with battery modules, however then we go back to cost vs benefit. The SSD brings the response time and queue length to a good level even when starting the 16 WUs at the same time upon restart.