Keep in mind that a single host can cause such a high poller runtime. Take a look at your processes running every polling cycle. You will notice that most probably only a single polling process will remain while others have been completed pretty fast. Try to identify that host and change its polling parameters.
This will become much easier with Cacti 1.0.
Boost itself it is great if you need to decrease I/O waits caused by slow disks or the amount of data you have to update. (In my case I have more than 400GB of RRDfiles.)
12/06/2016 08:22:51 AM - SYSTEM THOLD STATS: CPUTime:874 MaxRuntime:279 Tholds:135841 TotalHosts:2253 DownHosts:75 NewDownHosts:0 Processes: 14 completed, 4 running, 0 broken
12/06/2016 08:22:42 AM - SYSTEM STATS: Time:160.7313 Method:spine Processes:16 Threads:40 Hosts:2254 HostsPerProcess:141 DataSources:480698 RRDsProcessed:0
RRDUpdates:6193882 TotalTime:504 get_records:67.76 results_cycle:420.57 rrd_filename_and_template:128.21 rrd_lastupdate:22.34 rrdupdate:203.71 delete:7.64 timer_overhead:~6