Regression with two or three predictors

First lets construct different models:
> g <- lm(sav ~ p15 + p75 + inc)
> g1 <- lm(sav ~ p15 + p75)
> g2 <- lm(sav ~ p15 + inc)
Lets look at these models:
> summary(g,cor=F)

Call: lm(formula = sav ~ p15 + p75 + inc)
Residuals:
    Min     1Q  Median   3Q   Max
 -8.646 -2.567 -0.1192 2.28 10.37

Coefficients:
               Value Std. Error  t value Pr(>|t|)
(Intercept)  31.4581   7.4822     4.2044   0.0001
        p15  -0.4922   0.1490    -3.3021   0.0019
        p75  -1.5677   1.1208    -1.3988   0.1686
        inc  -0.0008   0.0009    -0.8938   0.3761

Residual standard error: 3.939 on 46 degrees of freedom
Multiple R-Squared: 0.2744
F-statistic: 5.797 on 3 and 46 degrees of freedom, the p-value is 0.001898
> summary(g1,cor=F)

Call: lm(formula = sav ~ p15 + p75)
Residuals:
    Min     1Q  Median    3Q   Max
 -8.725 -2.704 -0.1199 2.282 10.32

Coefficients:
               Value Std. Error  t value Pr(>|t|)
(Intercept)  30.6284   7.4085     4.1342   0.0001
        p15  -0.4709   0.1468    -3.2072   0.0024
        p75  -1.9342   1.0409    -1.8582   0.0694

Residual standard error: 3.931 on 47 degrees of freedom
Multiple R-Squared: 0.2618
F-statistic: 8.332 on 2 and 47 degrees of freedom, the p-value is 0.0007993
> summary(g2,cor=F)

Call: lm(formula = sav ~ p15 + inc)
Residuals:
    Min     1Q   Median    3Q   Max
 -8.117 -2.656 -0.00547 1.484 10.98

Coefficients:
               Value Std. Error  t value Pr(>|t|)
(Intercept)  22.7131   4.1523     5.4700   0.0000
        p15  -0.3303   0.0949    -3.4800   0.0011
        inc  -0.0013   0.0009    -1.4949   0.1416

Residual standard error: 3.979 on 47 degrees of freedom
Multiple R-Squared: 0.2435
F-statistic: 7.564 on 2 and 47 degrees of freedom, the p-value is 0.001419
Now lets see what is the correlation between different variables:
> r1<-g1$residual
> r2<-g2$residual
> cor(r1,p15)
[1] -5.5567e-17
> cor(r2,p15)
[1] -3.252871e-17
> cor(r2,r1)
[1] 0.9613393
What is 0.9613393? (note r1 and r2 are residuals).
Lets plot it:
> motif()
> brush(cbind(p15,p75,inc,r1,r2))
Play with it, rotate, choose different variables.  Click "quit" when you are done.
Now lets look at sum of squares:
>r<-g$residual
> sum(r^2)
[1] 713.7616
> sum(r1^2)
[1] 726.158
> sum(r2^2)
[1] 744.1213
Why first number is smallest?  Why second number is smaller than third?  What does it mean?
Let's look at sectional sum of squares:
> anova(g)
Analysis of Variance Table

Response: sav

Terms added sequentially (first to last)
          Df Sum of Sq  Mean Sq  F Value     Pr(F)
      p15  1  204.1241 204.1241 13.15524 0.0007158
      p75  1   53.3462  53.3462  3.43802 0.0701299
      inc  1   12.3964  12.3964  0.79892 0.3760707
Residuals 46  713.7616  15.5166
> anova(g1)
Analysis of Variance Table

Response: sav

Terms added sequentially (first to last)
          Df Sum of Sq  Mean Sq  F Value     Pr(F)
      p15  1  204.1241 204.1241 13.21177 0.0006877
      p75  1   53.3462  53.3462  3.45279 0.0694148
Residuals 47  726.1580  15.4502
> anova(g2)
Analysis of Variance Table

Response: sav

Terms added sequentially (first to last)
          Df Sum of Sq  Mean Sq  F Value     Pr(F)
      p15  1  204.1241 204.1241 12.89283 0.0007855
      inc  1   35.3829  35.3829  2.23484 0.1416147
Residuals 47  744.1213  15.8324                
What is the difference between RSS of the first and second model?  Do you see it some were, or do you need to calculate it?
Why SS(p75) are the same in both models, but SS(inc) are not?  Why SS(p15) is always the same?