My same problems was focused on notifications. A lot of administrators created a big amount of notifications, configured with a delay of 1 Min. As a result :
When SEPM was started, the console was running quite good with normal response time.
After a few minutes (about 20 Min) all the plannified repots beegin to run slowly because the first one need a long time, so the second one was ran later, and the third one .....
So after 20 Min, all the jvm memory was taken by the report processing.
The first solution was to delete half of these reports.
Then we ran an architecture divied into 3 SEPM and 1 SQL server:
1 SEPM manage clients connections
1 SEPM is used for management only
1 SEPM i used for reporting and notifications only
1 SQL 2008 R2 server
Each SEPM can get the task of another as a backup task.
Maybe that can help you.
Cheers