I could be wrong, but I'd think that global models like the UKMO will have some large biases in surface temperature in circumstances where you have a high airmass temperature but some cloud around. Model resolution likely isn't high enough to resolve many of the breaks in the cloud, and you'll end up with a more uniform layer of cloud being modelled than we would see in reality.
With 20C+ 850hPa temperatures and strong solar heating, the temperature would respond very quickly in cloud breaks, and this is probably isn't going to be captured in the model.