The best you can realistically accomplish is essentially balancing connections across the three connections. You'd have three network address translation (NAT) gateways (..could be in one physical box). The network would send a given individual network connection out one of the three circuits and the translated address would serve as the mechanism to assure return traffic came back to the right place.
The effectiveness / uniformity of distribution of traffic across these links is a statistical exercise. Your network device will choose the outbound path based on some collection of attributes (source / destination IP address, source/destination L4 port, source destination hardware address, etc, etc). The number of attributes used and the actual even distribution of these attributes determine how many connections go to a given link. It's entirely possible that an inefficient setup could send all of your workstation's traffic out a single link - which basically nets you no more performance than having purchased a single 2G connection. It's also possible (though less likely) that the amount of actual bandwidth you need is distributed amongst flows that are perfectly balanced across the three circuits.
In more sophisticated multihomed setups there would be additional routing information available to send traffic via a (nominally) better path, but the focus in those circumstances tend to be more on increasing availability than performance (although both can be positively affected). This also implies a lot of additional network design questions and a not inconsiderable degree of cost and paperwork.
Here are some issues worth considering-
2.) Unless you've got a decent degree of sophistication in your setup (i.e. the routing information I alluded to above) then there are a lot of failure modes that could be very troubling. Imagine that your address translation device didn't realize that one of the three connections was out of service. An unknown portion of traffic would be lost, but the impression to the end user would be randomly freezing sessions or odd performance while standard diagnostic tools (i.e. pings) might look just fine. Bear in mind that there are a lot of "gray" failures - which is to say the link might appear fine to the router but in reality no traffic can pass.
3.) The statistical connection sharing tends to work better with larger numbers of users and connections. If it's just you and a few others, there won't be as much of a win.
4.) None of the above addresses any sort of inbound connections (i.e. web server, mail server, etc). That's a whole other set of issues...
There are ways around these problems - special gateways, custom scripts, etc but they're typically not simple to maintain and troubleshoot. The underlying point is that nothing comes for free. You can get somewhat better performance under certain circumstances with multiple connections, but only at the cost of some amount of operational complexity (...which has a dollar value all its own) or you can bite the bullet and spend the additional money for a smaller number of larger connections.