While I don't have an expert answer, I can draft some thoughts on your first question.
To estimate one's security level you have to define what your adversaries' capabilities are (technical skills, financial power, time and number of men), plus other factors (why would they spend their resources on targeting you, ...). That will form your threat model.
Your set up does protect you from one specific attack configuration you mentioned : interception is made at compromised entry node and compromised exit node.
But that's not the only way to do traffic correlation. What if the attacker monitors your traffic when it is between your computer and your ISP? And bear in mind that correlation doesn't need to be perfect (extent knowledge of every packet content you sent and to whom) to be fatal : just knowing the fact that you're using Tor at a particular time can be enough (c.f. Bomb threats on a US campus to escape exams, only one guy connected to Tor at the time an email came from Tor. Fail).
So yes you're protected from some current attacks but not from lots of other.
Although of course, for someone to make all the investigation efforts, you would need to be valuable enough of a target.
Now, what about new attacks possibilities this set up would create ? Once again it depends on your threat model.
If someone deems it to be worth it, he can compromise your server, either remotely or physically. If you're not an expert at computer security, then sadly so your server will be hackable. Secure data centers can do a little more to prevent physical intrusions but everyone has a price and/or employees can be tricked.
And even if they couldn't compromise your server, the fact that you are alone on your machine doesn't guarantee its confidentiality, as analysing the power consumption of the host computer could leak information. OK, I reckon I added this one to freak you out, but this is truly possible and at least it shows that, provided someone is determined enough, you can't even imagine to what length they will go to beat you.
So in the end, it all comes down to your threat level. From who you want to conceal your activities and why ? You have to work with assumptions because technical facts are overwhelmingly unlimited.
What I wrote until now is objective, verifiable and commonly accepted facts, to the best of my knowledge.
But as no practical solution seem to emerge from these, I will now give my personal opinion which is to take with maximum precaution. I don't have enough cryptographic knowledge to assess the correctness of what I propose.
From all the threats I mentioned, the one that would worry me the most (because it is very likely to happen as it is cheap though most effective) is your ISP eavesdropping. Weather it is legal or not, weather they even know about it or not doesn't matter. (You could even enlarges this threat zone to the whole path between your computer and your server input).
This is serious because it is a single point of failure.
What I would do is obscuring the traffic that you send to your entry node. The fact that it is encrypted conceal the content but not its shape because Tor does not randomize traffic (a variable amount of junk bytes added to each packet). I think that using a VPN to connect to your server would address this issue. As the data would be decrypted on the server right before going into the node, the traffic you sent to the entry node does not leak anymore, and the extra layer of encryption makes statistical analysis harder.
Once again this is an hypothesis that need to be reviewed by cryptographers and confronted by security experts to assess its strength.
But I think that if there is a solution to statistical analysis of the Tor network, it may ressembles something like that. It does lack this extra twist that cryptographers add to a protocol, an elegant trick of which no regular people would have thought to, that turns a seemingly good idea into a piece of art.
