Large window sizes only help when there is absolutely no packet loss (which is not the Internet)
and even then it takes a long time to scale up.
When packet loss is significant, it will indicate congestion or a network issue and be a more important factor than latency.
Packet loss and latency are both significant. The recovery time from a single packet loss is proportional to the latency.
Large windows still help. Packet losses are usually less than 0.01%. Anything above approximately .1% is a strong indication of a possible issue.
Large windows should still help; although, there might be some rare situations where they make performance worse (if the network they are on is saturated and the extra traffic pushes it past a tipping point so packet loss increases).
When one is dealing with large windows, even 0.01% packet loss hurts. This is because it takes longer to get back up to full speed after a packet loss. It will still be faster than a smaller window size, but the packet loss will eat away more overall throughput than it would with a smaller window size.
Often websites are bypassing slow start by starting with an expanded initial congestion window.
Yes, if they are following the RFCs, they can use up to 4 packets (assuming the MTU is small) for the initial congestion window. That cuts two RTTs off the ramp up time (very roughly 10%). The last time I checked, web sites were usually transferring less than 100K bytes -- The use of 4 packets for the initial congestion window should actually help them a lot more than having a large window size (that they will never actually ramp up to). Where the large window sizes really help are large uni-directional transfers (ftp, sftp, etc).