Real world test: Sysbench MySQL benchmark
Another very common server role is a MySQL database server. The database will prefer the ganged mode or the unganged one? Let's see, starting with a simple (read only) test and four concurrent client threads:
The ganged mode is about 2% faster. However, remember that such a low difference can be explained by a measurement error.
How about the 256 thread test?
This time the unganged mode wins, but as the difference is under 1% I will consider round a tie.
Now, let's examine the more complex, transactional (read / write) scenario:
With four threads, the unganged mode seems to have a rather low 2% advantage, so I will consider this test a tie.
In the 512 thread test the unganged mode give us a <1% speedup. So, we have hardly any news: it was a tie.