I did some tests with DSL Reports and a couple mobile web browsers. The results were pretty interesting.

I used, Blazer 3.0 (natch), Xiino 3.2E, and ReqWireless 3.5. All the tests were as follows:
Network: GSM/ATTWS (duh!)
Location: Hollywood, FL
Specifics: Treo 600-1.13-AWS Hardware B Firmware 3.06
Signal strength: 4 bars
Battery: 100% charged
Time: Late morning (about 1130)

The tests are simple: downloading a set filesize. The server computes your time to download, latency, and computes an overall speed. File sizes are 5k, 50k, 100k, and 200k. Between each test, the cache was cleared. That being stated, here are my results.

Blazer:
5k file - unstable results (ie: too small)

50k file
32kbit/sec
2.155 latency
14.636 d/l time

100k file
33kbit/sec
2.123 latency
25.717 d/l time

200k
49kbit/sec
3.337 latency
35.172 d/l time

Xiino 3.2E build 2004-0909-17

5k and 50k, unstable results ???

100k
80kbit/sec !!!
3.8 latency
13.554 d/l time

200k
200kbit/sec
3.417 latency
11.395 d/l time

Now, ReqWireless 3.5

5k
7kbit/sec
0.115 latency
5.476 d/l time

50k
49kbit/sec
0.144 latency
8.191d/l time

100k
117kbit/sec
2.563 latency
6.922 d/l time

200k
181kbit/sec
0.342 latency
8.954 d/l time

I'm inclined to believe at "top" speed of 40-50kbit/sec, based on what results I've seen from other users with ATTWS. Naturally they were all using Blazer, but several people have noticed increased browsing speed with ReqWireless. Only downside is that you need JavaVM installed, a large-ish 1.5mg PDB. However, I prefer Xiino. Smaller file, visible toolbar, and an equally "real-time-fast" experience.

YMMV