The paper you link to is referring to 'Deep Web' meaning unindexed data, and it's claims may be roughly correct in this sense - at the time it was published. It's not referring to the Tor based network of hidden sites you also mention, however. The term DarkNet is often used for these sites, so for clarity, I will refer to 'Deep Web' when discussing normal web content that is unindexed, and DarkNet when referring to encrypted, hidden, anonymised websites.
Deep Web as unindexed pages
The paper you refer to is actually from 2001, and also seems to be a founding basis for the Wikipedia page on Deep Web. Because the paper is so old now, the likelihood of it's estimates being still correct, or even of the proportions of information being correct is in my opinion, remote. Opinions aside, the paper is very out of date, and a more up to date reference would be needed to estimate the size of indexed vs non indexed data on the web.
Anonymised proxy delivered websites, like The Silk Road and it's ilk, are the sites delivered over Tor and similar relays using obscure urls like https://silkroad3dff81.onion (deliberately not the actual Silk Road URL, although even the real one would be useless now it has been closed). These are more correctly known as DarkNets, or Dark Internet, although the terminology gets a bit mixed up.
Size of Darknet
I've looked through as much as I possibly can in the time I have on how big the DarkNet actually is, that is to say, how much data on the internet is stored in DarkNets. The search has been rather problematic, as most articles that are most prominent on the internet are actually quoting the 7.5TB or a reduction of this to a percentage, multiple etc. and this came from the orignal 2001 paper, you refer to, which can't be referring to Tor based Darknets, as it was published one year before Tor was launched.
The only practical way to measure the DarkNet, because the information is encrypted whilst passed, is by network analysis, which will only ever be a rough estimate (much encrypted data will not be DarkNet traffic). This IEEE paper covers an approach to measuring the DarkNet, but does not offer an actual figure for either bandwidth usage, or data stored.
The inherent problem of this question is that you are asking the size of hidden and secret data. To really size it, you'd need to be able to see it all, and even if you could, you couldn't know that there wasn't more information you were unable to see. So the answer is that the estimates given by journalist constantly are probably wrong, the size is probably changing all the time, and the size of the DarkNet is basically indeterminate.
Also of interest is this paper, which refers to the technical plausibility of growth of the DarkNet, concluding that growth is very possible, and likely.