It's been a while, but I think the broadcast queue is applied to the main interface & the calculation is based on the number of DLCIs/sub-interfaces on that interface.
From the
Cisco - Configuring and Troubleshooting Frame Relay
web page that I saved in 1999
A broadcast queue is given a maximum transmission rate (throughput) limit measured in bytes per second and packets per second. The queue is serviced to ensure that only this maximum is provided. The broadcast queue has priority when transmitting at a rate below the configured maximum, and hence has a guaranteed minimum bandwidth allocation. The two transmission rate limits are intended to avoid flooding the interface with broadcasts. The actual limit in any second is the first rate limit that is reached. Given the transmission rate restriction, additional buffering is required to store broadcast packets. The broadcast queue is configurable to store large numbers of broadcast packets. The queue size should be set to avoid loss of broadcast routing update packets. The exact size depends on the protocol being used and the number of packets required for each update. To be safe, the queue size should be set so that one complete routing update from each protocol and for each data-link connection identifier (DLCI) can be stored. As a general rule, start with 20 packets per DLCI. The byte rate should be less than both of the following:
N/4 times the minimum remote access rate (measured in bytes per second), where N is the number of DLCIs to which the broadcast must be replicated
1/4 the local access rate (measured in bytes per second)
The packet rate is not critical if the byte rate is set conservatively. In general, the packet rate should be set assuming 250-byte packets. The defaults are 64 queue size, 256,000 bytes per second (2,048,000 bps), and 36 pps.
