1. Why do larger mining pools have less orphans, assuming most miners even small ones are connected to the relay network?
Because it greatly reduces the time it takes to transmit blocks but does not completely eliminate it-- nothing can (due to the speed of light). In particular, something I didn't know until my conversation with them on July 4th: the nearest relay network hub to F2Pool is still 200ms away due to insane routing that sends traffic between some networks in china and singapore via the US (thanks NSA?).
2. Even if mining pools set higher fees, aren't the unconfirmed TX's still added to their mempools?
No.
3. How is it that 1MB just "happened" to be the magic number at which blocks are deemed to be "large" ?
I don't know what you're talking about there. AFAICT F2Pool would also consider e.g. 750k "large".
Do you mean why was 1MB selected as the particular hard limit in the protocol? ::shrugs:: It happens to be the the highest value you could sync over a modem and stay up with the network (though not for mining, due to latency), though that could be by chance.