As long as you implement PBR in hardware, it shouldn't affect the CPU.
Per the documentation:
http://www.cisco.com/en/US/docs/routers/7600/ios/15S/configuration/guide/layer3.html
The Policy Feature Card (PFC) and any Distributed Feature Cards (DFCs) provide hardware support for policy-based routing (PBR) for route-map sequences that use the match ip address, set ip next-hop, and ip default next-hop PBR keywords.
When configuring PBR, follow these guidelines and restrictions:
–The PFC provides hardware support for PBR configured on a tunnel interface.
–The PFC does not provide hardware support for PBR configured with the set ip next-hop keywords if the next hop is a tunnel interface.
–If the MSFC address falls within the range of a PBR ACL, traffic addressed to the MSFC is policy routed in hardware instead of being forwarded to the MSFC. To prevent policy routing of traffic addressed to the MSFC, configure PBR ACLs to deny traffic addressed to the MSFC.
–Any options in Cisco IOS ACLs that provide filtering in a PBR route-map that would cause flows to be sent to the MSFC to be switched in software are ignored. For example, logging is not supported in ACEs in Cisco IOS ACLs that provide filtering in PBR route-maps.
–PBR traffic through switching module ports where PBR is configured is routed in software if the switching module resets. (CSCee92191)
–In Cisco IOS Release 15.2(1)S1 and later releases, for efficient use of hardware resources, enter the platform ipv4 pbr optimize tcam command in global configuration mode when configuring multiple PBR sequences (or a single PBR sequence with multiple ACLs) in which more than one PBR ACL contains DENY entries.
_____________
Switches do not perform per-packet load balancing, it's done per-flow using CEF.
