While it is difficult to speculate without knowing more about your environment, I suspect that the CPU consumption is primarily driven by the forwarders, not the indexing.
You should be able to split the logs (for instance, from local clusters) to the nearest SMS, be it a standby or active, with other, designated for contingency logging. This said, if one of them will go down, the load on the remaining server will double and cause some unpleasantness.
If you indeed are forwarding to a six different SIEMs/log dumps, perhaps the solution is to have open source ETL do the split and forward from a single exported instance.