The delay was not due to the polling delay. It's currently set to poll every minute and I gave the job at least 5 minutes before assuming the trigger wasn't going to fire each time I investigated the problem.
When things run smoothly, I can drop a file into the network share, wait (for up to one minute) and the job will trigger and complete successfully. However, following a bad change to task sequencing, the job failed. I resolved the sequencing problem, moved the file that triggered the job to a temp folder then moved it back into the monitored folder. The trigger never fired. Without any changes, I then copied a differently named file into the monitored folder and the trigger fired (within a minute).
Regarding the use of only polling Samba folders - I believe non-polling file triggers can be unreliable when pointing to network drives, i.e. network interruptions can stop the trigger firing. Polling allows retries on error etc and therefore provides more resilience. For this reason, I'm happy to take the hit of a minor delay before the trigger fires.
Neill