You have misunderstood how linear models are parametrized in R. If you try fitting a linear model yourself in R, using lm(y~group) say, you will see that the claims you are making are incorrect.
Here is a little example:
> y <- c(1,2,3,5,6,7)
> group <- factor(c(1,1,1,2,2,2))
> fit <- lm(y~group)
> coefficients(summary(fit))
Estimate Std. Error t value Pr(>|t|)
(Intercept) 2 0.577 3.46 0.02572
group2 4 0.816 4.90 0.00805
You see that the intercept is the mean of the first group (mean of 1,2,3) and the second coefficient is the difference between the means of the two groups.
We can change the contrast parametrization to get the classical parametrization where the intercept is the grand mean:
> contrasts(group) <- contr.sum(levels(group))
> fit <- lm(y~group)
> coefficients(summary(fit))
Estimate Std. Error t value Pr(>|t|)
(Intercept) 4 0.408 9.8 0.000608
group1 -2 0.408 -4.9 0.008050
On the other hand, we can go back to the default parametrization in R for which the intercept is the mean of the first group:
> contrasts(group) <- contr.treatment(levels(group))
> fit <- lm(y~group)
> coefficients(summary(fit))
Estimate Std. Error t value Pr(>|t|)
(Intercept) 2 0.577 3.46 0.02572
group2 4 0.816 4.90 0.00805
To get the so-called "cell means" parameterization, we would remove the intercept:
> fit <- lm(y~0+group)
> coefficients(summary(fit))
Estimate Std. Error t value Pr(>|t|)
group1 2 0.577 3.46 0.025721
group2 6 0.577 10.39 0.000484
There are other possible parametrizations available in R: contr.sum and contr.treatment are not the only possibilities. Apart from the cell means model, the parametrizations are all equivalent in that they give the same anova table.
It is not clear what you mean by a "treatment effect model" because that is not a standard term in statistics. In R, the "treatment" contrast parametrization is the default whereby the intercept is the fitted value for the first group and the other coefficients are relative to the first group. In Stata software, a "treatment effects model" is a special purpose model for binary data, see:
http://www.stata.com/products/stb/journals/stb55.pdf
Obviously that doesn't have a lot of relevance to oneway anova.
The intercept item should not bet he average of one treatment. It should be the overall mean of all the treatment. There are two traditional model in ANOVA, one is called cell mean model and the other is treatment effect model.....