I have however not been able to come up with a way to abuse this and was hoping someone can point me in the right direction. Specifically, the way I see it, the checksum prevents me from forging a signature even if two different messages were signed with the same key. Why is this not the case?
Let us take a rather simplified example; consider the case where there is a single WOTS digit used to express the hash (and therefore a single WOTS digit to express the checksum); for this example, we'll have $W=16$.
The first message we sign is the hash value 2; that means that we publish $H^2(x)$ (where $x$ is from the private key), along with the checksum 14, which we publish as $H^{14}(y)$ (where $y$ is also from the private key)
Now, we sign (with the same private key) the hash value 13; that means we publish $H^{13}(x)$ and the checksum $H^3(y)$.
At this point, the attacker has enough information to generate a forgery for (say) the hash value 7. To do that, he'd take the $H^2(x)$ value from the first signature (which we'll call $a$) and compute $H^5(a)$; he'd take the $H^3(y)$ from the second signature (which we'll call $b$) and compute $H^6(b)$. The pair $H^5(a), H^6(b)$ is equal to $H^7(x), H^9(y)$, and so is a valid signature for 7, even though the attacker has no idea what the values for $x$ and $y$ are.
This attack extends easily to the real WOTS system (where a message is expressed in multiple digits), and the modification of WOTS+ (which stirs in a unique value for each hash invocation) doesn't actually make the attacker's job any harder.