Thanks Tufale, nice way to achieve this also. In fact, in our case, the job can not run more than once in a period of 15 minutes, but must run normally every hour, as it feeds a production system the guys on the shop floor needs to produce.
That's why we have this extra test on last run, so if it's more than 45 minutes, we allow the job to run anyway, to avoid to have to do any manual update of the shared memory flag (important in a 24/7 system 
)