Stephane, I believe that this parameter is careful not to override any value set by the config, so it should be possible to change it in the normal manner (ie "lctl set_param -P".
Also, since this only affects the in-memory cache, there is no problem if it is not set in the "ideal" manner for a second or two after mount while the OSS is applying the config parameters from the MGS.
Since we have very little information available to the filesystem (it was a challenge even getting "rotational" tone set consistently for the storage), I don't think it is practical for Lustre to auto-tune this based on the specific storage type.
Would this be best handled by describing the various options in the Lustre Operations Manual?
Andreas, thanks for your reply, that makes sense. With servers' memory performance increasing and probably the Linux page cache's performance also improving (not sure of that), it might be worth adding a note in the manual.
Today the Lustre manual says:
By default, read cache is enabled (read_cache_enable=1) for HDD OSDs and automatically disabled for flash OSDs (nonrotational=1).
Perhaps adding something like that would be a good general guidance:
However, depending on the hardware (performance of flash storage, type of server), it might be still beneficial to enable read_cache_enable with flash OSD. This can be done permanently with lctl set_param -P ....
Patches welcome 
You can also use the LU ticket to submit a patch to the manual, no need for a separate LUDOC ticket.