Egalisation of trace length. All signals see the same electrical length to avoid skew.
Singe this machine runs so fast and has two channels the data needs to arrive at exactly the same time. Any delay of a signal in respect to another will be seen as a glitch.
Dave: this kind of machinery is easiest opened of you take the front panel off. Look at the sides of the frontpanel ( short sides ) . There will be two screws per side. Take them out. The entire fornt pulls forward. The bnc connectors will stay as they are screwed in the chassis. Once the front is off you will find two or three screws and that entire metal plate with fans will lift out.
I have seen similar constructions on my 81101 as well as on the 54831 scopes.
The reson for the air douct is noise. They could have gotten away by putting 4 or 5 little fans simply vertical with a bit of plastic ducting but then you get the mosquito problem... Wheeeeeeeee. Little fans sound like mosquitos. So agilent solves that by having biiiiig fans and a bit of metal.
The germany plant ( boeblingen) has traditionally been their signal generator design house. Over the years they have come up with a number of really kick-ass designs. The most notably ones were the 8112 and 8116, which, for their time, were top notch and in production for two decades... After the three asics went obsolete they released the 31120 and many a customer complained that there was no alternative for the 8116... Yeah you got arb but not 50MHz... And not the crazy stuff like burst and slope control.
Used 8116 sold for more money than they had cost new. If you had a setup with one and it died , there was no alternative.
Even today, 10 years after last production the 8116 still catches a high price on ebay.