Brian Wakem has also replied

Brian Wakem has replied:

Forcing a filesystem cache of the table is probably loading the index into
memory too. In the second case you are not, so the extra time is probably
spent reading the index from disk, not the data file.

and we have replied him as soon:

But I am not loading the full table in cache, just only the corresponding MYD file as follows:

# cat /var/lib/mysql/full_text_investigations/*.MYD > /dev/null

The increment in cache size obtained through the "free" command matches with the MYD file size.

Thence, we can be sure that the query needs to read the MYD file, in order to explain the penalty in the benchmarks when not caching the file.

Thanks.