Hello Mannen med ljaaen, I can't answer your question about the GPU estimate but can about Uploading and Ready to Report.

If the client has internet access it will upload a workunit (WU) as soon as it completes. As you've observed once uploaded the status is changed as Ready to Report. The next time the client communicates with the server it will report the completed and uploaded WU.
When the client communicates with the server is based on the size of your cache and amount of work remaining to be completed. Net is it will communicate when it wants more work, or if there are WUs Ready to Report every 24 hours, which ever comes first.

Mannen med ljaaen,
Assuming you have the necessary network bandwidth, sending or receiving files is a very basic and easy task for a computer. In addition, grouping such operations does not save much, if at all.

On the contrary updating a large direct access database is a very expensive operation for a computer with lots of things which need to be done to ensure the integrity of the transaction itself and that of the database. And chaining such direct access transactions is a great opportunity to reduce the global stress of this activity.

Let's talk numbers now.
On a good day WCG gets 800,000 WUs done.
That means recording the completion of these 800,000 WUs plus recording the distribution of 800,000 new ones.
If these two operations can be done at the same time you have reduced the daily stress by close to 50 %.

And with all the multicores available nowadays it is not exceptional that several completed WUs are reported at once when the client asks for one new WU. And then the savings become really tremendous.

Future releases of the BOINC client will have new algorithms in order to differ work requests until it is possible to request several new ones at once, which will reduce the stress on the main BOINC database one step further.