HTL transient histograms are slightly faster than comparable HBOOK ones, even though they provide the user with more accurate information about bin centers.
HTL persistent histograms are somewhat slower than their transient counterparts. It would be possible, using more advanced Objectivity/DB techniques such as fast-filling methods and pinning in memory, to make them almost as fast as transient ones. However, a more practical approach in most cases is the use of transient histograms for repetitive operations such as filling, combined with persistent copies of the filled histogram for storage.
When raw performance is absolutely crucial, faster HTL histograms are also available. However, they do not provide filling time statistics.
Since HTL allows you to mix persistent and transient histograms, the optimal strategy when performance is an issue is to book and fill transient histogram and save them at the end of the run using the persistent histogram copy constructors (as explained in chapter Chapter 2., HTL at a glance).