I suggest, never delete data. Don't say if the tables is longer than 1000 truncate the end of the table. There needs to be real business logic in your plan like how long has this user been inactive. For example, if it is longer than 1 year then put them in a different table. You would have this happen weekly or monthly in a maintenance script in the middle of a slow time.
When you run into to many rows in your table then you should start sharding the tables or partitioning and put old data in old tables by year such as users_2011_jan, users_2011_feb or use numbers for the month. Then change your programming to work with this model. Maybe make a new table with less information to summarize the data in less columns and then only refer to the bigger partitioned tables when you need more information such as when the user is viewing their profile. All of this should be considered very carefully so in the future it isn't too expensive to re-factor. You could also put only the users which comes to your site all the time in one table and the users that never come in an archived set of tables.