Until a forwarding table is built (by observing the inbound source MAC addresses), any destination MAC address that hasn't been learned must be flooded (NOT BROADCAST).
That's the way bridges and L2 switches (which are basically multi-port bridges)work.
The difference between a flood and broadcast is that a broadcast has a destination address of all ones (ff.ff.ff.ff.ff.ff) where a flodded MAC will retain the original destination MAC (so that the intended recipient can recognize that it's for him).
Clearing the ARP cache is an IP thing, above Layer 2 ... Layer 3 switches still have to perform Layer 2 functionality ... so perhaps clearing the ARP cache also cleared the forwarding table.
A fresh table on a busy network can absolutely drive utilization to the roof ... I think you are seeing normal behavior. If it peaks initially and drops off, that'd be normal. If it hit the roof and stayed high, then there's a problem (like a loop).
Good Luck
Scott