If you are too aggressive with setting maintenance_work_mem, you may discover that autovacuum periodically causes out of memory errors on your system. You can typically identify that situation because the amount of memory listed in the logs for the allocation failure attempt will match that database setting.
Don't forget that up to autovacuum_max_workers can be active, each using that much memory. When estimating the available RAM on the system, it's better to err on the small side initially with this setting. Ramp it up after you've done enough monitoring to know what the peak memory usage on your server is.
