Those are good numbers. Especially without HyperServer and with just one executable. But it all depends on how you manage datasets and application logic on the server side.
The smaller the datasets resident in memory (coming from queries, tables or other) the less resources are consumed (memory and cpu), always having to multiply everything by the number of active users.
And clearly, the less "useless" traffic is generated between client and server, the better resources are managed.
I hope that with the management of "disconnected" datasets, as promised by Farshad for new lightweight DBgrids, much of the "in-memory persistent" data resources can be managed by shifting the dataset load to the client, freeing up resources on the server.
This should further improve Unigui's already excellent performances.