Scaled Dirichlet does not integrate to 1

I am not sure if it should, but when I scale a Dirichlet non-homogeneously, it seems to no longer integrate to 1:

import numpy as np
import tensorflow_probability as tfp
tfd = tfp.distributions
tfb = tfp.bijectors

# scale = [2.0, 2.0]  # Integrates to 1
scale = [2.0, 3.0]  # Does not integrate to 1

scaled_dir = tfd.TransformedDistribution(
  distribution=tfd.Dirichlet([4.0, 4.0], force_probs_to_zero_outside_support=True),
  bijector=tfb.Scale(scale),
)
x = np.linspace(0., scale[0], 10_000)
y = scale[1] - (scale[1]/scale[0])*x
np.testing.assert_allclose(x/scale[0] + y/scale[1], 1)
pts = np.concatenate((x[..., None], y[..., None]), -1)

integral = trapezoid(np.exp(scaled_dir.log_prob(pts)), x=x)
np.testing.assert_allclose(integral, 1, rtol=1e-6)  # Fails

AssertionError: 
Not equal to tolerance rtol=1e-06, atol=0

Mismatched elements: 1 / 1 (100%)
Max absolute difference: 0.21553527
Max relative difference: 0.21553527
 x: array(0.784465)
 y: array(1)