I am using SQL Server 2005 replication with anonymous subscriptions and web
synchronization. The server is SQL 2005 Standard and the subscribers are SQL
2005 Express. Every subscriber syncs with the publisher once an hour. There
are no re-publishers.

The subscription databases each run a client program that works on the local
subset of data (partitioning is done by SUSER_SNAME).

My problem is that periodically the client program will fail with an
Identity Range Check Constraint. The error can be fixed by replicating with
the server.

The way I understand SQL2005 replication with SQL2005 subscribers, each
subscriber should be getting a primary and a secondary range of identity
values. When the primary range runs out they should switch to the secondary
range and refresh the primary range (and vice-versa when the secondary range
runs out).

Alternatively, non-SQL2005 subscribers should request a new identity range
when their identity range threshold is exceeded. I have a subscriber identity
range of 10000 and a threshold of 80%. None of the subscribers have inserted
2000 rows between replications (they typically insert no more than 100 rows).

Please help.