There are more issues than the GPIB transfer speeds. If you send a GPIB command to start measuring, the instrument will have to take the reading before having anything to return back. If the multimeter is able to do 10 readings/second, each reading would then take 100ms.

Some instruments are built to be very quick. Some instruments are built to be slow, but having huge resolution. I just want to recommend that you measure the actual time needed for one GPIB measurement cycle before trying to change the baudrate, something that should only allow you to gain 4ms if you have an infinite baudrate and no processing time.

Looking at the serial communication may be required, if the microcontroller or the PC has a lot of lag in the communication. But then it isn't the baudrate that is the problem, but how fast the two sides will react to received data. If the PC is receiving the data, and configured to not respond to data until the receive buffer is full, or 100ms has passed, then the serial link will seriously affect your total cycle time.