Yes we want it to scale, even if it means full nodes run only by large organizations. The benefits of decentralization would be very real and tangible even if only, say, 1000 organizations (companies, universities, etc) were able to host and operate Bitcoin nodes. This would still make Bitcoin clearly decentralized compared to a single company (like MasterCard) processing all your CC transactions. For example Mastercard decided of their own accord to block CC donations to Wikileaks [1] but one of these 1000 hypothetical organizations running full nodes would be unable to do such a thing and block specific transactions.
The number of full nodes decreased in large part because in the early history of Bitcoin in late 2010 or early 2011 your only option to have a secure wallet or to mine reliably was to run a full node locally. But nowadays there are many alternative lightweight clients using the SPV protocol, and a lot of reliable mining pools (it was in mid-2011 that pools started mining more coins than solo miners running full nodes). So of course many users stopped running full nodes.
I wasn't commenting on what is better or worse. I was simply stating that there is a tangible effect on the decentralization of the network when you have a limited number of nodes. For example, maintaining privacy becomes much harder as these nodes become major "pipelines" for fresh first hop transactions. The list goes on, but in reality it depends on how many there are, a number we can't predict. FWIW, I think your number is generous.
The number of full nodes decreased in large part because in the early history of Bitcoin in late 2010 or early 2011 your only option to have a secure wallet or to mine reliably was to run a full node locally. But nowadays there are many alternative lightweight clients using the SPV protocol, and a lot of reliable mining pools (it was in mid-2011 that pools started mining more coins than solo miners running full nodes). So of course many users stopped running full nodes.
[1] http://www.theregister.co.uk/2010/12/07/wikileaks_latest/