+There are any number of race conditions that may result from the fact
+that account names are not unique over time. Moreover, it's a bad
+idea to perform lengthy operations while holding the database lock.
+In order to deal with both of these problems, we use a worker thread
+for each account name that ever exists. On 32-bit systems with large
+numbers of accounts, this may cause the NM process to run out of
+*virtual* memory! This problem may be remedied by decreasing the
+maximum stack size.