As the name says: ID is an (unique) identifier. I think it's better to explain your customer that you are working on different projects than having dublicate (custom) ID's?

Thats true, but it is hard, to tell a customer that different projects share the ID workspace, especially if the customer does not trust your methods. They may think, you are hiding information (bugs with IDs of other projects) for them.

Its similar, if you dont ship a version 0.19 of mantis, but a version 0.20. So everyone asks, why 0.19 is missing (maybe its too buggy for realize). It always leaves a bad taste.

by counter are you looking for say, select MAX(value) where customfieldid = 10 for example, or a unique identifier.
i.e.
a counter where you take the previous max+1
or a counter where the counter will always be unique (i.e. it cant have been used + deleted)

The first version would be fairly easy to implement as the value could be looked up at runtime. However, using this method, and as mantis doesn't use transactions, it would always be possible that you could somehow duplicate an id if it was entered at the same time.

The last one.

Counter should be unique but deletion is allowed and leaves a hole. The deleted id should not be used anymore. Its the same like the id for reports (autoincrement).

gbvfghf