In addition to what Jonathan mentioned, The design matrix specified in the the two cases is as follows

Y = beta*X+epsilon

and in once case, X is simply A B C and in other case X is A Apm B Bpm C Cpm
Firstly, the parametric odulator should not be highly correlated with the condition itself. You can look at both the design matrices and there should be row below those design matrices (when you explore design) that mentions contrast / parameter estimability. This should be ideally white. Gray is still okay, but if you see that, it could be one of the explanations of the results you observe. 

Second, The brain activation in the second design matrix is distributed to condition and parametric modulation. As i mentioned earlier the same contrast i.e. differences between conditions A and B should not be severly affected if your parametric modulator is orthogonal to the conditions. This is rarely the case. When the modulator is highly correlated then the activation is shown in the regressor contrast instead of the condition contrast. 

By extracting the timecourse of activation, you can independently verify your claim that the activation is related to the parametric modulator.